You asked ChatGPT the question your homepage claims to answer. It named four other companies. Yours was not one of them.
The model is not ignoring you. It has no basis to choose you.
That distinction decides what you do next. One version of this problem is technical and gets fixed in an afternoon. The other version is positioning, and no amount of publishing will touch it.
I checked which version is more common. On October 9, 2026 I ran a study of 232 B2B homepages. Of 169 headlines, 157 could be swapped onto a competitor’s site without becoming false. That is 93%.
The technical side was fine in that sample. All 204 readable sites let in all six AI search crawlers. 203 served their homepage text before JavaScript ran.
A larger crawl of 3,877 B2B software sites found the same shape with a caveat. Only 35 of 3,517 readable robots.txt files blocked a search bot, but 34 sites blocked ChatGPT at the firewall while their robots.txt invited it in.
So access does fail, on roughly one site in a hundred. The sentences fail on nine in ten.
How do you see your brand visibility in ChatGPT?
Ask five times, in five separate conversations, logged out.
One check tells you nothing. SparkToro found fewer than one in a hundred answer pairs name the same list of brands. Kevin Indig found only 2.2% of citations survive three runs of the same prompt. Ask once and you have a coin flip.
Run it properly instead. Open a temporary chat. Ask the category question a buyer would ask, without your company name in it. Record who gets named. Close the conversation and repeat four more times.
Stay logged out. A signed in account carries your own history into the answer, which is how people convince themselves they are visible when they are not.
The full method is in how to track brand mentions in AI search. For this article, five runs on one prompt is enough to see the pattern.
Three outcomes, three different problems
Three outcomes, three different problems.
You appear in most runs. Fine. Check the other nine prompts before you celebrate, because category questions behave differently than the one you happened to try.
You appear on your brand name and nowhere else. The engines know you exist. They do not consider you an answer to anything. That is positioning.
You appear nowhere, including on your own brand name. The engines probably cannot reach or read your site. That is technical, and it is the cheaper problem to have. It is also the rarer one, and it usually sits in the firewall rather than in robots.txt.
Could a competitor publish your homepage?
Take your primary claim. Ask whether a direct competitor could put that exact sentence on their site without it becoming false.
Most claims fail this test. In my sample, 105 of 169 headlines (62%) would work on any direct competitor’s site. Another 44 (26%) would work for any company in any industry. Twelve headlines out of 169 described only the company that published them.
Call it the adoptable claim. It reads like positioning, it survives every internal review, and it gives a retrieval system nothing to select on. When the engine assembles a list of vendors, it needs a reason to pick one company over another. Your sentence gave it none.
The specifics were missing in predictable places. Half the headlines, 84 of 169, said what the company sells. Only 21 said who it is for. Only 17 contained a detail a reader could check. And 69 of 169 answered neither question, which is 41% of homepages telling a buyer nothing they could act on.
Score every claim on your homepage three ways:
- Adoptable. A competitor could say it unchanged. The claim does no work.
- Contested. Several companies claim it. It differentiates only if you can prove it.
- Defensible. True of you and not of the field, and a reader could verify it.
Do this with the competitors’ actual sentences open in another tab. Scoring your claims against your impression of the category is how everyone talks themselves into a passing grade.
Two notes on how I ran that study, because the method decides what the number is worth. The 232 companies were marketing leaders in my own LinkedIn network, so it is a convenience sample rather than a random one. And the swap ratings came from AI raters. Two runs of the same model agreed on 155 of 169. A second model from another company agreed on 88%. Under a two-of-three vote, 163 of 169 came out not unique. I report the looser figure of 157.
Can a machine even read your claim?
Open your homepage, view the source, and search for your headline sentence.
If it is not in the initial HTML, most engines never see it. A Vercel and MERJ analysis found that of the major AI crawlers, only Google’s and Apple’s execute JavaScript. ChatGPT’s crawler fetches JavaScript files and runs none of them. Only Google’s and Apple’s crawlers render.
This produces the cheapest fix in the whole exercise. A defensible claim sitting in a hero image, a JavaScript carousel, or a tabbed panel scores zero for retrieval no matter how good it is. Move it into static HTML and the engine can finally read the one sentence that would have earned you the mention.
A common version: a named framework rendered as three single words in a row, with no sentence around them and no company name attached. The position is defensible. The machine sees three adjectives.
The answer sets the shortlist
Because the answer sets the shortlist, and the shortlist gets validated by a human later.
Gartner found 45% of B2B buyers used generative AI during a recent purchase. Of those, 69% then asked a sales rep to validate what the AI told them.
Your competitor gets named, the buyer brings that name to a call, and the rep confirms it. You never enter the process. The pipeline effect shows up a quarter later as fewer inbound conversations, and nothing in your analytics explains why.
Fix it in this order
Work in order. Access, then rendering, then the claim.
- Confirm retrieval crawlers can reach your pages.
- Confirm your primary claim sits in the initial HTML.
- Score every claim for adoptability against real competitor sentences.
- Rewrite the claims that failed, then put the survivors in your title tag, meta description and Open Graph card.
That order matters. Improving brand visibility in AI search follows the same sequence. Fix a sentence on a page an engine cannot fetch and you change nothing you can measure.
One more step people skip. Write down the competitors the engines named in your place, with counts. That list tells you which companies the model considers interchangeable with you. Read their homepage sentences next to yours. The gap between them is the work.
Name the ground nobody else claims
Find the claims that zero competitors make. Then check whether any of them appear in your title tag, your meta description, or your Open Graph card.
Usually none of them do. Companies bury the one true thing below the fold, under a headline any competitor could have written. That is a sequencing problem, and it is cheap to fix.
The share cards are worse than the headlines. Across 194 sites with a meta description, 74 never named the company. Twenty-four share-card titles started with the word “Home,” and 13 of those sat on WordPress sites. A machine reading that card learns nothing about who published the page.
Ask before you publish another post
Take one question into your next content meeting. If a competitor copied this page word for word, would anything on it become false?
If the answer is no, more posts will not help. You will publish ten more pages any competitor could have written, and the engines will keep naming them.
The audit answers both halves at once. It scores your claims against quoted competitor sentences. It checks whether machines can reach and read them. It runs 150 observations across ChatGPT, Google AI Mode and Perplexity to show who gets named in your place.
Scott King helps B2B technology companies build AI marketing systems and create differentiated content.